Jailbreaking Bad: The AI Industry is Cooked

Welcome back to The FAIK Files! When tech gets weird, we're here to help make sense of it all.

In this week's show:

• We explore how randomness in AI systems creates the illusion of thought
• A disturbing case of AI chatbots being weaponized for cyberstalking
• Anthropic's new approach to preventing AI jailbreaks
• And our AI dumpster fire of the week is... AI... the whole thing... all of it

Subscribe to our BRAND NEW YouTube channel! We'll be adding a wide variety of content shortly, but would love your help building the subscriber base up before we release our first video. You can find the channel at: https://www.youtube.com/@theFAIKfiles

Want to leave us a voicemail? Here's the magic link to do just that: https://sayhi.chat/FAIK

You can also join our Discord server here: https://discord.gg/cThqEnMhJz

*** NOTES AND REFERENCES ***

Randomness in AI Systems:

• Overview of deterministic vs stochastic systems in AI
• Understanding temperature settings in LLMs
• How diffusion models use random seeds
• Discussion of parameters like top-K and top-P
• Relationship between randomness and perceived intelligence
• Lots of good overviews available at the Prompt Engineering Guide website: https://www.promptingguide.ai/introduction/settings

AI-Enabled Cyberstalking:

• The Guardian: Stalking AI Chatbot Impersonator
• Case study of James Florence's 7-year cyberstalking campaign
• Discussion of platforms CrushOn.ai and JanitorAI
• Implications for future harassment scenarios

Anthropic's Constitutional Classifiers:

• Anthropic's post: Constitutional Classifiers: Defending against universal jailbreaks
• Anthropic's demo website: https://claude.ai/constitutional-classifiers
• Details of 3,000+ hours of red-teaming with 405 participants
• System architecture and implementation
• Success rate: 95% of jailbreak attempts blocked
• Only 23.7% inference overhead

AI Dumpster Fire -- The entire AI industry:

• Inconsistent naming conventions across companies
• Bad and inconsistent public relations strategies
• Arms race between US and China
• Environmental and ethical concerns
• And more...

*** THE BOILERPLATE ***

About The FAIK Files:

The FAIK Files is an offshoot project from Perry Carpenter's most recent book, FAIK: A Practical Guide to Living in a World of Deepfakes, Disinformation, and AI-Generated Deceptions.

• Get the Book: FAIK: A Practical Guide to Living in a World of Deepfakes, Disinformation, and AI-Generated Deceptions (Amazon Associates link)
• Check out the website for more info: https://thisbookisfaik.com

Check out Perry & Mason's other show, the Digital Folklore Podcast:

• Apple Podcasts: https://podcasts.apple.com/us/podcast/digital-folklore/id1657374458
• Spotify: https://open.spotify.com/show/2v1BelkrbSRSkHEP4cYffj?si=u4XTTY4pR4qEqh5zMNSVQA
• Other: https://digitalfolklore.fm 

Want to connect with us? Here's how:

Connect with Perry:

• Perry on LinkedIn: https://www.linkedin.com/in/perrycarpenter
• Perry on X: https://x.com/perrycarpenter
• Perry on BlueSky: https://bsky.app/profile/perrycarpenter.bsky.social

Connect with Mason:

• Mason on LinkedIn: https://www.linkedin.com/in/mason-amadeus-a853a7242/
• Mason on BlueSky: https://bsky.app/profile/pregnantsonic.com